Add the new ND6_IFF_ACCEPT_RTADV flag to control whether to accept
[dragonfly.git] / sys / netinet6 / nd6.c
1 /*      $FreeBSD: src/sys/netinet6/nd6.c,v 1.2.2.15 2003/05/06 06:46:58 suz Exp $       */
2 /*      $DragonFly: src/sys/netinet6/nd6.c,v 1.13 2004/12/30 02:26:12 hsu Exp $ */
3 /*      $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $   */
4
5 /*
6  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. Neither the name of the project nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33
34 /*
35  * XXX
36  * KAME 970409 note:
37  * BSD/OS version heavily modifies this code, related to llinfo.
38  * Since we don't have BSD/OS version of net/route.c in our hand,
39  * I left the code mostly as it was in 970310.  -- itojun
40  */
41
42 #include "opt_inet.h"
43 #include "opt_inet6.h"
44
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/callout.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/socket.h>
51 #include <sys/sockio.h>
52 #include <sys/time.h>
53 #include <sys/kernel.h>
54 #include <sys/protosw.h>
55 #include <sys/errno.h>
56 #include <sys/syslog.h>
57 #include <sys/queue.h>
58 #include <sys/sysctl.h>
59
60 #include <net/if.h>
61 #include <net/if_dl.h>
62 #include <net/if_types.h>
63 #include <net/if_atm.h>
64 #include <net/route.h>
65
66 #include <netinet/in.h>
67 #include <netinet/if_ether.h>
68 #include <netinet/if_fddi.h>
69 #include <netinet6/in6_var.h>
70 #include <netinet/ip6.h>
71 #include <netinet6/ip6_var.h>
72 #include <netinet6/nd6.h>
73 #include <netinet6/in6_prefix.h>
74 #include <netinet/icmp6.h>
75
76 #include "use_loop.h"
77
78 #include <net/net_osdep.h>
79
80 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
81 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
82
83 #define SIN6(s) ((struct sockaddr_in6 *)s)
84 #define SDL(s) ((struct sockaddr_dl *)s)
85
86 /* timer values */
87 int     nd6_prune       = 1;    /* walk list every 1 seconds */
88 int     nd6_delay       = 5;    /* delay first probe time 5 second */
89 int     nd6_umaxtries   = 3;    /* maximum unicast query */
90 int     nd6_mmaxtries   = 3;    /* maximum multicast query */
91 int     nd6_useloopback = 1;    /* use loopback interface for local traffic */
92 int     nd6_gctimer     = (60 * 60 * 24); /* 1 day: garbage collection timer */
93
94 /* preventing too many loops in ND option parsing */
95 int nd6_maxndopt = 10;  /* max # of ND options allowed */
96
97 int nd6_maxnudhint = 0; /* max # of subsequent upper layer hints */
98
99 #ifdef ND6_DEBUG
100 int nd6_debug = 1;
101 #else
102 int nd6_debug = 0;
103 #endif
104
105 /* for debugging? */
106 static int nd6_inuse, nd6_allocated;
107
108 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
109 static size_t nd_ifinfo_indexlim = 8;
110 struct nd_ifinfo *nd_ifinfo = NULL;
111 struct nd_drhead nd_defrouter;
112 struct nd_prhead nd_prefix = { 0 };
113
114 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
115 static struct sockaddr_in6 all1_sa;
116
117 static void nd6_slowtimo (void *);
118 static int regen_tmpaddr (struct in6_ifaddr *);
119
120 struct callout nd6_slowtimo_ch;
121 struct callout nd6_timer_ch;
122 extern struct callout in6_tmpaddrtimer_ch;
123
124 void
125 nd6_init(void)
126 {
127         static int nd6_init_done = 0;
128         int i;
129
130         if (nd6_init_done) {
131                 log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
132                 return;
133         }
134
135         all1_sa.sin6_family = AF_INET6;
136         all1_sa.sin6_len = sizeof(struct sockaddr_in6);
137         for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
138                 all1_sa.sin6_addr.s6_addr[i] = 0xff;
139
140         /* initialization of the default router list */
141         TAILQ_INIT(&nd_defrouter);
142
143         nd6_init_done = 1;
144
145         /* start timer */
146         callout_init(&nd6_slowtimo_ch);
147         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
148             nd6_slowtimo, NULL);
149 }
150
151 void
152 nd6_ifattach(struct ifnet *ifp)
153 {
154
155         /*
156          * We have some arrays that should be indexed by if_index.
157          * since if_index will grow dynamically, they should grow too.
158          */
159         if (nd_ifinfo == NULL || if_index >= nd_ifinfo_indexlim) {
160                 size_t n;
161                 caddr_t q;
162
163                 while (if_index >= nd_ifinfo_indexlim)
164                         nd_ifinfo_indexlim <<= 1;
165
166                 /* grow nd_ifinfo */
167                 n = nd_ifinfo_indexlim * sizeof(struct nd_ifinfo);
168                 q = (caddr_t)malloc(n, M_IP6NDP, M_WAITOK);
169                 bzero(q, n);
170                 if (nd_ifinfo) {
171                         bcopy((caddr_t)nd_ifinfo, q, n/2);
172                         free((caddr_t)nd_ifinfo, M_IP6NDP);
173                 }
174                 nd_ifinfo = (struct nd_ifinfo *)q;
175         }
176
177 #define ND nd_ifinfo[ifp->if_index]
178
179         /*
180          * Don't initialize if called twice.
181          * XXX: to detect this, we should choose a member that is never set
182          * before initialization of the ND structure itself.  We formaly used
183          * the linkmtu member, which was not suitable because it could be 
184          * initialized via "ifconfig mtu".
185          */
186         if (ND.basereachable)
187                 return;
188
189         ND.linkmtu = ifindex2ifnet[ifp->if_index]->if_mtu;
190         ND.chlim = IPV6_DEFHLIM;
191         ND.basereachable = REACHABLE_TIME;
192         ND.reachable = ND_COMPUTE_RTIME(ND.basereachable);
193         ND.retrans = RETRANS_TIMER;
194         ND.receivedra = 0;
195         /*
196          * Note that the default value of ip6_accept_rtadv is 0, which means
197          * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
198          * here.
199          */
200         ND.flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
201         nd6_setmtu(ifp);
202 #undef ND
203 }
204
205 /*
206  * Reset ND level link MTU. This function is called when the physical MTU
207  * changes, which means we might have to adjust the ND level MTU.
208  */
209 void
210 nd6_setmtu(struct ifnet *ifp)
211 {
212         struct nd_ifinfo *ndi = &nd_ifinfo[ifp->if_index];
213         u_long oldmaxmtu = ndi->maxmtu;
214         u_long oldlinkmtu = ndi->linkmtu;
215
216         switch (ifp->if_type) {
217         case IFT_ARCNET:        /* XXX MTU handling needs more work */
218                 ndi->maxmtu = MIN(60480, ifp->if_mtu);
219                 break;
220         case IFT_ETHER:
221                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
222                 break;
223         case IFT_FDDI:
224                 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
225                 break;
226         case IFT_ATM:
227                 ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
228                 break;
229         case IFT_IEEE1394:      /* XXX should be IEEE1394MTU(1500) */
230                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
231                 break;
232 #ifdef IFT_IEEE80211
233         case IFT_IEEE80211:     /* XXX should be IEEE80211MTU(1500) */
234                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
235                 break;
236 #endif
237         default:
238                 ndi->maxmtu = ifp->if_mtu;
239                 break;
240         }
241
242         if (oldmaxmtu != ndi->maxmtu) {
243                 /*
244                  * If the ND level MTU is not set yet, or if the maxmtu
245                  * is reset to a smaller value than the ND level MTU,
246                  * also reset the ND level MTU.
247                  */
248                 if (ndi->linkmtu == 0 ||
249                     ndi->maxmtu < ndi->linkmtu) {
250                         ndi->linkmtu = ndi->maxmtu;
251                         /* also adjust in6_maxmtu if necessary. */
252                         if (oldlinkmtu == 0) {
253                                 /*
254                                  * XXX: the case analysis is grotty, but
255                                  * it is not efficient to call in6_setmaxmtu()
256                                  * here when we are during the initialization
257                                  * procedure.
258                                  */
259                                 if (in6_maxmtu < ndi->linkmtu)
260                                         in6_maxmtu = ndi->linkmtu;
261                         } else
262                                 in6_setmaxmtu();
263                 }
264         }
265 #undef MIN
266 }
267
268 void
269 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
270 {
271         bzero(ndopts, sizeof(*ndopts));
272         ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
273         ndopts->nd_opts_last
274                 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
275
276         if (icmp6len == 0) {
277                 ndopts->nd_opts_done = 1;
278                 ndopts->nd_opts_search = NULL;
279         }
280 }
281
282 /*
283  * Take one ND option.
284  */
285 struct nd_opt_hdr *
286 nd6_option(union nd_opts *ndopts)
287 {
288         struct nd_opt_hdr *nd_opt;
289         int olen;
290
291         if (!ndopts)
292                 panic("ndopts == NULL in nd6_option");
293         if (!ndopts->nd_opts_last)
294                 panic("uninitialized ndopts in nd6_option");
295         if (!ndopts->nd_opts_search)
296                 return NULL;
297         if (ndopts->nd_opts_done)
298                 return NULL;
299
300         nd_opt = ndopts->nd_opts_search;
301
302         /* make sure nd_opt_len is inside the buffer */
303         if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
304                 bzero(ndopts, sizeof(*ndopts));
305                 return NULL;
306         }
307
308         olen = nd_opt->nd_opt_len << 3;
309         if (olen == 0) {
310                 /*
311                  * Message validation requires that all included
312                  * options have a length that is greater than zero.
313                  */
314                 bzero(ndopts, sizeof(*ndopts));
315                 return NULL;
316         }
317
318         ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
319         if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
320                 /* option overruns the end of buffer, invalid */
321                 bzero(ndopts, sizeof(*ndopts));
322                 return NULL;
323         } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
324                 /* reached the end of options chain */
325                 ndopts->nd_opts_done = 1;
326                 ndopts->nd_opts_search = NULL;
327         }
328         return nd_opt;
329 }
330
331 /*
332  * Parse multiple ND options.
333  * This function is much easier to use, for ND routines that do not need
334  * multiple options of the same type.
335  */
336 int
337 nd6_options(union nd_opts *ndopts)
338 {
339         struct nd_opt_hdr *nd_opt;
340         int i = 0;
341
342         if (!ndopts)
343                 panic("ndopts == NULL in nd6_options");
344         if (!ndopts->nd_opts_last)
345                 panic("uninitialized ndopts in nd6_options");
346         if (!ndopts->nd_opts_search)
347                 return 0;
348
349         while (1) {
350                 nd_opt = nd6_option(ndopts);
351                 if (!nd_opt && !ndopts->nd_opts_last) {
352                         /*
353                          * Message validation requires that all included
354                          * options have a length that is greater than zero.
355                          */
356                         icmp6stat.icp6s_nd_badopt++;
357                         bzero(ndopts, sizeof(*ndopts));
358                         return -1;
359                 }
360
361                 if (!nd_opt)
362                         goto skip1;
363
364                 switch (nd_opt->nd_opt_type) {
365                 case ND_OPT_SOURCE_LINKADDR:
366                 case ND_OPT_TARGET_LINKADDR:
367                 case ND_OPT_MTU:
368                 case ND_OPT_REDIRECTED_HEADER:
369                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
370                                 nd6log((LOG_INFO,
371                                     "duplicated ND6 option found (type=%d)\n",
372                                     nd_opt->nd_opt_type));
373                                 /* XXX bark? */
374                         } else {
375                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
376                                         = nd_opt;
377                         }
378                         break;
379                 case ND_OPT_PREFIX_INFORMATION:
380                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
381                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
382                                         = nd_opt;
383                         }
384                         ndopts->nd_opts_pi_end =
385                                 (struct nd_opt_prefix_info *)nd_opt;
386                         break;
387                 default:
388                         /*
389                          * Unknown options must be silently ignored,
390                          * to accomodate future extension to the protocol.
391                          */
392                         nd6log((LOG_DEBUG,
393                             "nd6_options: unsupported option %d - "
394                             "option ignored\n", nd_opt->nd_opt_type));
395                 }
396
397 skip1:
398                 i++;
399                 if (i > nd6_maxndopt) {
400                         icmp6stat.icp6s_nd_toomanyopt++;
401                         nd6log((LOG_INFO, "too many loop in nd opt\n"));
402                         break;
403                 }
404
405                 if (ndopts->nd_opts_done)
406                         break;
407         }
408
409         return 0;
410 }
411
412 /*
413  * ND6 timer routine to expire default route list and prefix list
414  */
415 void
416 nd6_timer(void *ignored_arg)
417 {
418         int s;
419         struct llinfo_nd6 *ln;
420         struct nd_defrouter *dr;
421         struct nd_prefix *pr;
422         struct ifnet *ifp;
423         struct in6_ifaddr *ia6, *nia6;
424         struct in6_addrlifetime *lt6;
425         
426         s = splnet();
427         callout_reset(&nd6_timer_ch, nd6_prune * hz,
428                       nd6_timer, NULL);
429
430         ln = llinfo_nd6.ln_next;
431         while (ln && ln != &llinfo_nd6) {
432                 struct rtentry *rt;
433                 struct sockaddr_in6 *dst;
434                 struct llinfo_nd6 *next = ln->ln_next;
435                 /* XXX: used for the DELAY case only: */
436                 struct nd_ifinfo *ndi = NULL;
437
438                 if ((rt = ln->ln_rt) == NULL) {
439                         ln = next;
440                         continue;
441                 }
442                 if ((ifp = rt->rt_ifp) == NULL) {
443                         ln = next;
444                         continue;
445                 }
446                 ndi = &nd_ifinfo[ifp->if_index];
447                 dst = (struct sockaddr_in6 *)rt_key(rt);
448
449                 if (ln->ln_expire > time_second) {
450                         ln = next;
451                         continue;
452                 }
453
454                 /* sanity check */
455                 if (!rt)
456                         panic("rt=0 in nd6_timer(ln=%p)", ln);
457                 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
458                         panic("rt_llinfo(%p) is not equal to ln(%p)",
459                               rt->rt_llinfo, ln);
460                 if (!dst)
461                         panic("dst=0 in nd6_timer(ln=%p)", ln);
462
463                 switch (ln->ln_state) {
464                 case ND6_LLINFO_INCOMPLETE:
465                         if (ln->ln_asked < nd6_mmaxtries) {
466                                 ln->ln_asked++;
467                                 ln->ln_expire = time_second +
468                                         nd_ifinfo[ifp->if_index].retrans / 1000;
469                                 nd6_ns_output(ifp, NULL, &dst->sin6_addr,
470                                         ln, 0);
471                         } else {
472                                 struct mbuf *m = ln->ln_hold;
473                                 if (m) {
474                                         if (rt->rt_ifp) {
475                                                 /*
476                                                  * Fake rcvif to make ICMP error
477                                                  * more helpful in diagnosing
478                                                  * for the receiver.
479                                                  * XXX: should we consider
480                                                  * older rcvif?
481                                                  */
482                                                 m->m_pkthdr.rcvif = rt->rt_ifp;
483                                         }
484                                         icmp6_error(m, ICMP6_DST_UNREACH,
485                                                     ICMP6_DST_UNREACH_ADDR, 0);
486                                         ln->ln_hold = NULL;
487                                 }
488                                 next = nd6_free(rt);
489                         }
490                         break;
491                 case ND6_LLINFO_REACHABLE:
492                         if (ln->ln_expire) {
493                                 ln->ln_state = ND6_LLINFO_STALE;
494                                 ln->ln_expire = time_second + nd6_gctimer;
495                         }
496                         break;
497
498                 case ND6_LLINFO_STALE:
499                         /* Garbage Collection(RFC 2461 5.3) */
500                         if (ln->ln_expire)
501                                 next = nd6_free(rt);
502                         break;
503
504                 case ND6_LLINFO_DELAY:
505                         if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
506                                 /* We need NUD */
507                                 ln->ln_asked = 1;
508                                 ln->ln_state = ND6_LLINFO_PROBE;
509                                 ln->ln_expire = time_second +
510                                         ndi->retrans / 1000;
511                                 nd6_ns_output(ifp, &dst->sin6_addr,
512                                               &dst->sin6_addr,
513                                               ln, 0);
514                         } else {
515                                 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
516                                 ln->ln_expire = time_second + nd6_gctimer;
517                         }
518                         break;
519                 case ND6_LLINFO_PROBE:
520                         if (ln->ln_asked < nd6_umaxtries) {
521                                 ln->ln_asked++;
522                                 ln->ln_expire = time_second +
523                                         nd_ifinfo[ifp->if_index].retrans / 1000;
524                                 nd6_ns_output(ifp, &dst->sin6_addr,
525                                                &dst->sin6_addr, ln, 0);
526                         } else {
527                                 next = nd6_free(rt);
528                         }
529                         break;
530                 }
531                 ln = next;
532         }
533         
534         /* expire default router list */
535         dr = TAILQ_FIRST(&nd_defrouter);
536         while (dr) {
537                 if (dr->expire && dr->expire < time_second) {
538                         struct nd_defrouter *t;
539                         t = TAILQ_NEXT(dr, dr_entry);
540                         defrtrlist_del(dr);
541                         dr = t;
542                 } else {
543                         dr = TAILQ_NEXT(dr, dr_entry);
544                 }
545         }
546
547         /*
548          * expire interface addresses.
549          * in the past the loop was inside prefix expiry processing.
550          * However, from a stricter speci-confrmance standpoint, we should
551          * rather separate address lifetimes and prefix lifetimes.
552          */
553   addrloop:
554         for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
555                 nia6 = ia6->ia_next;
556                 /* check address lifetime */
557                 lt6 = &ia6->ia6_lifetime;
558                 if (IFA6_IS_INVALID(ia6)) {
559                         int regen = 0;
560
561                         /*
562                          * If the expiring address is temporary, try
563                          * regenerating a new one.  This would be useful when
564                          * we suspended a laptop PC, then turned it on after a
565                          * period that could invalidate all temporary
566                          * addresses.  Although we may have to restart the
567                          * loop (see below), it must be after purging the
568                          * address.  Otherwise, we'd see an infinite loop of
569                          * regeneration. 
570                          */
571                         if (ip6_use_tempaddr &&
572                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
573                                 if (regen_tmpaddr(ia6) == 0)
574                                         regen = 1;
575                         }
576
577                         in6_purgeaddr(&ia6->ia_ifa);
578
579                         if (regen)
580                                 goto addrloop; /* XXX: see below */
581                 }
582                 if (IFA6_IS_DEPRECATED(ia6)) {
583                         int oldflags = ia6->ia6_flags;
584
585                         ia6->ia6_flags |= IN6_IFF_DEPRECATED;
586
587                         /*
588                          * If a temporary address has just become deprecated,
589                          * regenerate a new one if possible.
590                          */
591                         if (ip6_use_tempaddr &&
592                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
593                             (oldflags & IN6_IFF_DEPRECATED) == 0) {
594
595                                 if (regen_tmpaddr(ia6) == 0) {
596                                         /*
597                                          * A new temporary address is
598                                          * generated.
599                                          * XXX: this means the address chain
600                                          * has changed while we are still in
601                                          * the loop.  Although the change
602                                          * would not cause disaster (because
603                                          * it's not a deletion, but an
604                                          * addition,) we'd rather restart the
605                                          * loop just for safety.  Or does this 
606                                          * significantly reduce performance??
607                                          */
608                                         goto addrloop;
609                                 }
610                         }
611                 } else {
612                         /*
613                          * A new RA might have made a deprecated address
614                          * preferred.
615                          */
616                         ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
617                 }
618         }
619
620         /* expire prefix list */
621         pr = nd_prefix.lh_first;
622         while (pr) {
623                 /*
624                  * check prefix lifetime.
625                  * since pltime is just for autoconf, pltime processing for
626                  * prefix is not necessary.
627                  */
628                 if (pr->ndpr_expire && pr->ndpr_expire < time_second) {
629                         struct nd_prefix *t;
630                         t = pr->ndpr_next;
631
632                         /*
633                          * address expiration and prefix expiration are
634                          * separate.  NEVER perform in6_purgeaddr here.
635                          */
636
637                         prelist_remove(pr);
638                         pr = t;
639                 } else
640                         pr = pr->ndpr_next;
641         }
642         splx(s);
643 }
644
645 static int
646 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
647                                          address */
648 {
649         struct ifaddr *ifa;
650         struct ifnet *ifp;
651         struct in6_ifaddr *public_ifa6 = NULL;
652
653         ifp = ia6->ia_ifa.ifa_ifp;
654         TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_list) {
655                 struct in6_ifaddr *it6;
656
657                 if (ifa->ifa_addr->sa_family != AF_INET6)
658                         continue;
659
660                 it6 = (struct in6_ifaddr *)ifa;
661
662                 /* ignore no autoconf addresses. */
663                 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
664                         continue;
665
666                 /* ignore autoconf addresses with different prefixes. */
667                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
668                         continue;
669
670                 /*
671                  * Now we are looking at an autoconf address with the same
672                  * prefix as ours.  If the address is temporary and is still
673                  * preferred, do not create another one.  It would be rare, but
674                  * could happen, for example, when we resume a laptop PC after
675                  * a long period.
676                  */
677                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
678                     !IFA6_IS_DEPRECATED(it6)) {
679                         public_ifa6 = NULL;
680                         break;
681                 }
682
683                 /*
684                  * This is a public autoconf address that has the same prefix
685                  * as ours.  If it is preferred, keep it.  We can't break the
686                  * loop here, because there may be a still-preferred temporary
687                  * address with the prefix.
688                  */
689                 if (!IFA6_IS_DEPRECATED(it6))
690                     public_ifa6 = it6;
691         }
692
693         if (public_ifa6 != NULL) {
694                 int e;
695
696                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
697                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
698                             " tmp addr,errno=%d\n", e);
699                         return(-1);
700                 }
701                 return(0);
702         }
703
704         return(-1);
705 }
706
707 /*
708  * Nuke neighbor cache/prefix/default router management table, right before
709  * ifp goes away.
710  */
711 void
712 nd6_purge(struct ifnet *ifp)
713 {
714         struct llinfo_nd6 *ln, *nln;
715         struct nd_defrouter *dr, *ndr, drany;
716         struct nd_prefix *pr, *npr;
717
718         /* Nuke default router list entries toward ifp */
719         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
720                 /*
721                  * The first entry of the list may be stored in
722                  * the routing table, so we'll delete it later.
723                  */
724                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
725                         ndr = TAILQ_NEXT(dr, dr_entry);
726                         if (dr->ifp == ifp)
727                                 defrtrlist_del(dr);
728                 }
729                 dr = TAILQ_FIRST(&nd_defrouter);
730                 if (dr->ifp == ifp)
731                         defrtrlist_del(dr);
732         }
733
734         /* Nuke prefix list entries toward ifp */
735         for (pr = nd_prefix.lh_first; pr; pr = npr) {
736                 npr = pr->ndpr_next;
737                 if (pr->ndpr_ifp == ifp) {
738                         /*
739                          * Previously, pr->ndpr_addr is removed as well,
740                          * but I strongly believe we don't have to do it.
741                          * nd6_purge() is only called from in6_ifdetach(),
742                          * which removes all the associated interface addresses
743                          * by itself.
744                          * (jinmei@kame.net 20010129)
745                          */
746                         prelist_remove(pr);
747                 }
748         }
749
750         /* cancel default outgoing interface setting */
751         if (nd6_defifindex == ifp->if_index)
752                 nd6_setdefaultiface(0);
753
754         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
755                 /* refresh default router list */
756                 bzero(&drany, sizeof(drany));
757                 defrouter_delreq(&drany, 0);
758                 defrouter_select();
759         }
760
761         /*
762          * Nuke neighbor cache entries for the ifp.
763          * Note that rt->rt_ifp may not be the same as ifp,
764          * due to KAME goto ours hack.  See RTM_RESOLVE case in
765          * nd6_rtrequest(), and ip6_input().
766          */
767         ln = llinfo_nd6.ln_next;
768         while (ln && ln != &llinfo_nd6) {
769                 struct rtentry *rt;
770                 struct sockaddr_dl *sdl;
771
772                 nln = ln->ln_next;
773                 rt = ln->ln_rt;
774                 if (rt && rt->rt_gateway &&
775                     rt->rt_gateway->sa_family == AF_LINK) {
776                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
777                         if (sdl->sdl_index == ifp->if_index)
778                                 nln = nd6_free(rt);
779                 }
780                 ln = nln;
781         }
782 }
783
784 struct rtentry *
785 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
786 {
787         struct rtentry *rt;
788         struct sockaddr_in6 sin6;
789
790         bzero(&sin6, sizeof(sin6));
791         sin6.sin6_len = sizeof(struct sockaddr_in6);
792         sin6.sin6_family = AF_INET6;
793         sin6.sin6_addr = *addr6;
794 #ifdef SCOPEDROUTING
795         sin6.sin6_scope_id = in6_addr2scopeid(ifp, addr6);
796 #endif
797         rt = rtlookup((struct sockaddr *)&sin6, create, 0UL);
798         if (rt && !(rt->rt_flags & RTF_LLINFO)) {
799                 /*
800                  * This is the case for the default route.
801                  * If we want to create a neighbor cache for the address, we
802                  * should free the route for the destination and allocate an
803                  * interface route.
804                  */
805                 if (create) {
806                         --rt->rt_refcnt;
807                         rt = NULL;
808                 }
809         }
810         if (!rt) {
811                 if (create && ifp) {
812                         int e;
813
814                         /*
815                          * If no route is available and create is set,
816                          * we allocate a host route for the destination
817                          * and treat it like an interface route.
818                          * This hack is necessary for a neighbor which can't
819                          * be covered by our own prefix.
820                          */
821                         struct ifaddr *ifa =
822                                 ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
823                         if (ifa == NULL)
824                                 return(NULL);
825
826                         /*
827                          * Create a new route.  RTF_LLINFO is necessary
828                          * to create a Neighbor Cache entry for the
829                          * destination in nd6_rtrequest which will be
830                          * called in rtrequest via ifa->ifa_rtrequest.
831                          */
832                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
833                                            ifa->ifa_addr,
834                                            (struct sockaddr *)&all1_sa,
835                                            (ifa->ifa_flags |
836                                             RTF_HOST | RTF_LLINFO) &
837                                            ~RTF_CLONING,
838                                            &rt)) != 0)
839                                 log(LOG_ERR,
840                                     "nd6_lookup: failed to add route for a "
841                                     "neighbor(%s), errno=%d\n",
842                                     ip6_sprintf(addr6), e);
843                         if (rt == NULL)
844                                 return(NULL);
845                         if (rt->rt_llinfo) {
846                                 struct llinfo_nd6 *ln =
847                                         (struct llinfo_nd6 *)rt->rt_llinfo;
848                                 ln->ln_state = ND6_LLINFO_NOSTATE;
849                         }
850                 } else
851                         return(NULL);
852         }
853         rt->rt_refcnt--;
854         /*
855          * Validation for the entry.
856          * Note that the check for rt_llinfo is necessary because a cloned
857          * route from a parent route that has the L flag (e.g. the default
858          * route to a p2p interface) may have the flag, too, while the
859          * destination is not actually a neighbor.
860          * XXX: we can't use rt->rt_ifp to check for the interface, since
861          *      it might be the loopback interface if the entry is for our
862          *      own address on a non-loopback interface. Instead, we should
863          *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
864          *      interface.
865          */
866         if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
867             rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
868             (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
869                 if (create) {
870                         log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
871                             ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
872                         /* xxx more logs... kazu */
873                 }
874                 return(NULL);
875         }
876         return(rt);
877 }
878
879 /*
880  * Detect if a given IPv6 address identifies a neighbor on a given link.
881  * XXX: should take care of the destination of a p2p link?
882  */
883 int
884 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
885 {
886         struct ifaddr *ifa;
887         int i;
888
889 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
890 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
891
892         /*
893          * A link-local address is always a neighbor.
894          * XXX: we should use the sin6_scope_id field rather than the embedded
895          * interface index.
896          */
897         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
898             ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
899                 return(1);
900
901         /*
902          * If the address matches one of our addresses,
903          * it should be a neighbor.
904          */
905         TAILQ_FOREACH(ifa, &ifp->if_addrlist, ifa_list) {
906                 if (ifa->ifa_addr->sa_family != AF_INET6)
907                         next: continue;
908
909                 for (i = 0; i < 4; i++) {
910                         if ((IFADDR6(ifa).s6_addr32[i] ^
911                              addr->sin6_addr.s6_addr32[i]) &
912                             IFMASK6(ifa).s6_addr32[i])
913                                 goto next;
914                 }
915                 return(1);
916         }
917
918         /*
919          * Even if the address matches none of our addresses, it might be
920          * in the neighbor cache.
921          */
922         if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
923                 return(1);
924
925         return(0);
926 #undef IFADDR6
927 #undef IFMASK6
928 }
929
930 /*
931  * Free an nd6 llinfo entry.
932  */
933 struct llinfo_nd6 *
934 nd6_free(struct rtentry *rt)
935 {
936         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
937         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
938         struct nd_defrouter *dr;
939
940         /*
941          * we used to have pfctlinput(PRC_HOSTDEAD) here. 
942          * even though it is not harmful, it was not really necessary.
943          */
944
945         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
946                 int s;
947                 s = splnet();
948                 dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
949                                       rt->rt_ifp);
950
951                 if (ln->ln_router || dr) {
952                         /*
953                          * rt6_flush must be called whether or not the neighbor
954                          * is in the Default Router List.
955                          * See a corresponding comment in nd6_na_input().
956                          */
957                         rt6_flush(&in6, rt->rt_ifp);
958                 }
959
960                 if (dr) {
961                         /*
962                          * Unreachablity of a router might affect the default
963                          * router selection and on-link detection of advertised
964                          * prefixes.
965                          */
966
967                         /*
968                          * Temporarily fake the state to choose a new default
969                          * router and to perform on-link determination of
970                          * prefixes correctly.
971                          * Below the state will be set correctly,
972                          * or the entry itself will be deleted.
973                          */
974                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
975
976                         /*
977                          * Since defrouter_select() does not affect the
978                          * on-link determination and MIP6 needs the check
979                          * before the default router selection, we perform
980                          * the check now.
981                          */
982                         pfxlist_onlink_check();
983
984                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
985                                 /*
986                                  * It is used as the current default router,
987                                  * so we have to move it to the end of the
988                                  * list and choose a new one.
989                                  * XXX: it is not very efficient if this is
990                                  *      the only router.
991                                  */
992                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
993                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
994
995                                 defrouter_select();
996                         }
997                 }
998                 splx(s);
999         }
1000
1001         /*
1002          * Before deleting the entry, remember the next entry as the
1003          * return value.  We need this because pfxlist_onlink_check() above
1004          * might have freed other entries (particularly the old next entry) as
1005          * a side effect (XXX).
1006          */
1007         next = ln->ln_next;
1008
1009         /*
1010          * Detach the route from the routing tree and the list of neighbor
1011          * caches, and disable the route entry not to be used in already
1012          * cached routes.
1013          */
1014         rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
1015                   rt_mask(rt), 0, (struct rtentry **)0);
1016
1017         return(next);
1018 }
1019
1020 /*
1021  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1022  *
1023  * XXX cost-effective metods?
1024  */
1025 void
1026 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1027 {
1028         struct llinfo_nd6 *ln;
1029
1030         /*
1031          * If the caller specified "rt", use that.  Otherwise, resolve the
1032          * routing table by supplied "dst6".
1033          */
1034         if (!rt) {
1035                 if (!dst6)
1036                         return;
1037                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1038                         return;
1039         }
1040
1041         if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1042             (rt->rt_flags & RTF_LLINFO) == 0 ||
1043             !rt->rt_llinfo || !rt->rt_gateway ||
1044             rt->rt_gateway->sa_family != AF_LINK) {
1045                 /* This is not a host route. */
1046                 return;
1047         }
1048
1049         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1050         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1051                 return;
1052
1053         /*
1054          * if we get upper-layer reachability confirmation many times,
1055          * it is possible we have false information.
1056          */
1057         if (!force) {
1058                 ln->ln_byhint++;
1059                 if (ln->ln_byhint > nd6_maxnudhint)
1060                         return;
1061         }
1062
1063         ln->ln_state = ND6_LLINFO_REACHABLE;
1064         if (ln->ln_expire)
1065                 ln->ln_expire = time_second +
1066                         nd_ifinfo[rt->rt_ifp->if_index].reachable;
1067 }
1068
1069 void
1070 nd6_rtrequest(int req, struct rtentry *rt,
1071               struct rt_addrinfo *info) /* xxx unused */
1072 {
1073         struct sockaddr *gate = rt->rt_gateway;
1074         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1075         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1076         struct ifnet *ifp = rt->rt_ifp;
1077         struct ifaddr *ifa;
1078
1079         if ((rt->rt_flags & RTF_GATEWAY))
1080                 return;
1081
1082         if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1083                 /*
1084                  * This is probably an interface direct route for a link
1085                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1086                  * We do not need special treatment below for such a route.
1087                  * Moreover, the RTF_LLINFO flag which would be set below
1088                  * would annoy the ndp(8) command.
1089                  */
1090                 return;
1091         }
1092
1093         if (req == RTM_RESOLVE &&
1094             (nd6_need_cache(ifp) == 0 || /* stf case */
1095              !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1096                 /*
1097                  * FreeBSD and BSD/OS often make a cloned host route based
1098                  * on a less-specific route (e.g. the default route).
1099                  * If the less specific route does not have a "gateway"
1100                  * (this is the case when the route just goes to a p2p or an
1101                  * stf interface), we'll mistakenly make a neighbor cache for
1102                  * the host route, and will see strange neighbor solicitation
1103                  * for the corresponding destination.  In order to avoid the
1104                  * confusion, we check if the destination of the route is
1105                  * a neighbor in terms of neighbor discovery, and stop the
1106                  * process if not.  Additionally, we remove the LLINFO flag
1107                  * so that ndp(8) will not try to get the neighbor information
1108                  * of the destination.
1109                  */
1110                 rt->rt_flags &= ~RTF_LLINFO;
1111                 return;
1112         }
1113
1114         switch (req) {
1115         case RTM_ADD:
1116                 /*
1117                  * There is no backward compatibility :)
1118                  *
1119                  * if ((rt->rt_flags & RTF_HOST) == 0 &&
1120                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1121                  *         rt->rt_flags |= RTF_CLONING;
1122                  */
1123                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1124                         /*
1125                          * Case 1: This route should come from
1126                          * a route to interface.  RTF_LLINFO flag is set
1127                          * for a host route whose destination should be
1128                          * treated as on-link.
1129                          */
1130                         rt_setgate(rt, rt_key(rt),
1131                                    (struct sockaddr *)&null_sdl);
1132                         gate = rt->rt_gateway;
1133                         SDL(gate)->sdl_type = ifp->if_type;
1134                         SDL(gate)->sdl_index = ifp->if_index;
1135                         if (ln)
1136                                 ln->ln_expire = time_second;
1137 #if 1
1138                         if (ln && ln->ln_expire == 0) {
1139                                 /* kludge for desktops */
1140 #if 0
1141                                 printf("nd6_rtequest: time.tv_sec is zero; "
1142                                        "treat it as 1\n");
1143 #endif
1144                                 ln->ln_expire = 1;
1145                         }
1146 #endif
1147                         if ((rt->rt_flags & RTF_CLONING))
1148                                 break;
1149                 }
1150                 /*
1151                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1152                  * We don't do that here since llinfo is not ready yet.
1153                  *
1154                  * There are also couple of other things to be discussed:
1155                  * - unsolicited NA code needs improvement beforehand
1156                  * - RFC2461 says we MAY send multicast unsolicited NA
1157                  *   (7.2.6 paragraph 4), however, it also says that we
1158                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1159                  *   we don't have anything like it right now.
1160                  *   note that the mechanism needs a mutual agreement
1161                  *   between proxies, which means that we need to implement
1162                  *   a new protocol, or a new kludge.
1163                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1164                  *   we need to check ip6forwarding before sending it.
1165                  *   (or should we allow proxy ND configuration only for
1166                  *   routers?  there's no mention about proxy ND from hosts)
1167                  */
1168 #if 0
1169                 /* XXX it does not work */
1170                 if (rt->rt_flags & RTF_ANNOUNCE)
1171                         nd6_na_output(ifp,
1172                               &SIN6(rt_key(rt))->sin6_addr,
1173                               &SIN6(rt_key(rt))->sin6_addr,
1174                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1175                               1, NULL);
1176 #endif
1177                 /* FALLTHROUGH */
1178         case RTM_RESOLVE:
1179                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1180                         /*
1181                          * Address resolution isn't necessary for a point to
1182                          * point link, so we can skip this test for a p2p link.
1183                          */
1184                         if (gate->sa_family != AF_LINK ||
1185                             gate->sa_len < sizeof(null_sdl)) {
1186                                 log(LOG_DEBUG,
1187                                     "nd6_rtrequest: bad gateway value: %s\n",
1188                                     if_name(ifp));
1189                                 break;
1190                         }
1191                         SDL(gate)->sdl_type = ifp->if_type;
1192                         SDL(gate)->sdl_index = ifp->if_index;
1193                 }
1194                 if (ln != NULL)
1195                         break;  /* This happens on a route change */
1196                 /*
1197                  * Case 2: This route may come from cloning, or a manual route
1198                  * add with a LL address.
1199                  */
1200                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1201                 rt->rt_llinfo = (caddr_t)ln;
1202                 if (!ln) {
1203                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1204                         break;
1205                 }
1206                 nd6_inuse++;
1207                 nd6_allocated++;
1208                 bzero(ln, sizeof(*ln));
1209                 ln->ln_rt = rt;
1210                 /* this is required for "ndp" command. - shin */
1211                 if (req == RTM_ADD) {
1212                         /*
1213                          * gate should have some valid AF_LINK entry,
1214                          * and ln->ln_expire should have some lifetime
1215                          * which is specified by ndp command.
1216                          */
1217                         ln->ln_state = ND6_LLINFO_REACHABLE;
1218                         ln->ln_byhint = 0;
1219                 } else {
1220                         /*
1221                          * When req == RTM_RESOLVE, rt is created and
1222                          * initialized in rtrequest(), so rt_expire is 0.
1223                          */
1224                         ln->ln_state = ND6_LLINFO_NOSTATE;
1225                         ln->ln_expire = time_second;
1226                 }
1227                 rt->rt_flags |= RTF_LLINFO;
1228                 ln->ln_next = llinfo_nd6.ln_next;
1229                 llinfo_nd6.ln_next = ln;
1230                 ln->ln_prev = &llinfo_nd6;
1231                 ln->ln_next->ln_prev = ln;
1232
1233                 /*
1234                  * check if rt_key(rt) is one of my address assigned
1235                  * to the interface.
1236                  */
1237                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1238                                           &SIN6(rt_key(rt))->sin6_addr);
1239                 if (ifa) {
1240                         caddr_t macp = nd6_ifptomac(ifp);
1241                         ln->ln_expire = 0;
1242                         ln->ln_state = ND6_LLINFO_REACHABLE;
1243                         ln->ln_byhint = 0;
1244                         if (macp) {
1245                                 bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1246                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1247                         }
1248                         if (nd6_useloopback) {
1249                                 rt->rt_ifp = &loif[0];  /* XXX */
1250                                 /*
1251                                  * Make sure rt_ifa be equal to the ifaddr
1252                                  * corresponding to the address.
1253                                  * We need this because when we refer
1254                                  * rt_ifa->ia6_flags in ip6_input, we assume
1255                                  * that the rt_ifa points to the address instead
1256                                  * of the loopback address.
1257                                  */
1258                                 if (ifa != rt->rt_ifa) {
1259                                         IFAFREE(rt->rt_ifa);
1260                                         IFAREF(ifa);
1261                                         rt->rt_ifa = ifa;
1262                                 }
1263                         }
1264                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1265                         ln->ln_expire = 0;
1266                         ln->ln_state = ND6_LLINFO_REACHABLE;
1267                         ln->ln_byhint = 0;
1268
1269                         /* join solicited node multicast for proxy ND */
1270                         if (ifp->if_flags & IFF_MULTICAST) {
1271                                 struct in6_addr llsol;
1272                                 int error;
1273
1274                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1275                                 llsol.s6_addr16[0] = htons(0xff02);
1276                                 llsol.s6_addr16[1] = htons(ifp->if_index);
1277                                 llsol.s6_addr32[1] = 0;
1278                                 llsol.s6_addr32[2] = htonl(1);
1279                                 llsol.s6_addr8[12] = 0xff;
1280
1281                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1282                                         nd6log((LOG_ERR, "%s: failed to join "
1283                                             "%s (errno=%d)\n", if_name(ifp),
1284                                             ip6_sprintf(&llsol), error));
1285                                 }
1286                         }
1287                 }
1288                 break;
1289
1290         case RTM_DELETE:
1291                 if (!ln)
1292                         break;
1293                 /* leave from solicited node multicast for proxy ND */
1294                 if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1295                     (ifp->if_flags & IFF_MULTICAST) != 0) {
1296                         struct in6_addr llsol;
1297                         struct in6_multi *in6m;
1298
1299                         llsol = SIN6(rt_key(rt))->sin6_addr;
1300                         llsol.s6_addr16[0] = htons(0xff02);
1301                         llsol.s6_addr16[1] = htons(ifp->if_index);
1302                         llsol.s6_addr32[1] = 0;
1303                         llsol.s6_addr32[2] = htonl(1);
1304                         llsol.s6_addr8[12] = 0xff;
1305
1306                         IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1307                         if (in6m)
1308                                 in6_delmulti(in6m);
1309                 }
1310                 nd6_inuse--;
1311                 ln->ln_next->ln_prev = ln->ln_prev;
1312                 ln->ln_prev->ln_next = ln->ln_next;
1313                 ln->ln_prev = NULL;
1314                 rt->rt_llinfo = 0;
1315                 rt->rt_flags &= ~RTF_LLINFO;
1316                 if (ln->ln_hold)
1317                         m_freem(ln->ln_hold);
1318                 Free((caddr_t)ln);
1319         }
1320 }
1321
1322 int
1323 nd6_ioctl(u_long cmd, caddr_t   data, struct ifnet *ifp)
1324 {
1325         struct in6_drlist *drl = (struct in6_drlist *)data;
1326         struct in6_prlist *prl = (struct in6_prlist *)data;
1327         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1328         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1329         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1330         struct nd_defrouter *dr, any;
1331         struct nd_prefix *pr;
1332         struct rtentry *rt;
1333         int i = 0, error = 0;
1334         int s;
1335
1336         switch (cmd) {
1337         case SIOCGDRLST_IN6:
1338                 /*
1339                  * obsolete API, use sysctl under net.inet6.icmp6
1340                  */
1341                 bzero(drl, sizeof(*drl));
1342                 s = splnet();
1343                 dr = TAILQ_FIRST(&nd_defrouter);
1344                 while (dr && i < DRLSTSIZ) {
1345                         drl->defrouter[i].rtaddr = dr->rtaddr;
1346                         if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1347                                 /* XXX: need to this hack for KAME stack */
1348                                 drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1349                         } else
1350                                 log(LOG_ERR,
1351                                     "default router list contains a "
1352                                     "non-linklocal address(%s)\n",
1353                                     ip6_sprintf(&drl->defrouter[i].rtaddr));
1354
1355                         drl->defrouter[i].flags = dr->flags;
1356                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1357                         drl->defrouter[i].expire = dr->expire;
1358                         drl->defrouter[i].if_index = dr->ifp->if_index;
1359                         i++;
1360                         dr = TAILQ_NEXT(dr, dr_entry);
1361                 }
1362                 splx(s);
1363                 break;
1364         case SIOCGPRLST_IN6:
1365                 /*
1366                  * obsolete API, use sysctl under net.inet6.icmp6
1367                  */
1368                 /*
1369                  * XXX meaning of fields, especialy "raflags", is very
1370                  * differnet between RA prefix list and RR/static prefix list.
1371                  * how about separating ioctls into two?
1372                  */
1373                 bzero(prl, sizeof(*prl));
1374                 s = splnet();
1375                 pr = nd_prefix.lh_first;
1376                 while (pr && i < PRLSTSIZ) {
1377                         struct nd_pfxrouter *pfr;
1378                         int j;
1379
1380                         (void)in6_embedscope(&prl->prefix[i].prefix,
1381                             &pr->ndpr_prefix, NULL, NULL);
1382                         prl->prefix[i].raflags = pr->ndpr_raf;
1383                         prl->prefix[i].prefixlen = pr->ndpr_plen;
1384                         prl->prefix[i].vltime = pr->ndpr_vltime;
1385                         prl->prefix[i].pltime = pr->ndpr_pltime;
1386                         prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1387                         prl->prefix[i].expire = pr->ndpr_expire;
1388
1389                         pfr = pr->ndpr_advrtrs.lh_first;
1390                         j = 0;
1391                         while (pfr) {
1392                                 if (j < DRLSTSIZ) {
1393 #define RTRADDR prl->prefix[i].advrtr[j]
1394                                         RTRADDR = pfr->router->rtaddr;
1395                                         if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1396                                                 /* XXX: hack for KAME */
1397                                                 RTRADDR.s6_addr16[1] = 0;
1398                                         } else
1399                                                 log(LOG_ERR,
1400                                                     "a router(%s) advertises "
1401                                                     "a prefix with "
1402                                                     "non-link local address\n",
1403                                                     ip6_sprintf(&RTRADDR));
1404 #undef RTRADDR
1405                                 }
1406                                 j++;
1407                                 pfr = pfr->pfr_next;
1408                         }
1409                         prl->prefix[i].advrtrs = j;
1410                         prl->prefix[i].origin = PR_ORIG_RA;
1411
1412                         i++;
1413                         pr = pr->ndpr_next;
1414                 }
1415               {
1416                 struct rr_prefix *rpp;
1417
1418                 for (rpp = LIST_FIRST(&rr_prefix); rpp;
1419                      rpp = LIST_NEXT(rpp, rp_entry)) {
1420                         if (i >= PRLSTSIZ)
1421                                 break;
1422                         (void)in6_embedscope(&prl->prefix[i].prefix,
1423                             &pr->ndpr_prefix, NULL, NULL);
1424                         prl->prefix[i].raflags = rpp->rp_raf;
1425                         prl->prefix[i].prefixlen = rpp->rp_plen;
1426                         prl->prefix[i].vltime = rpp->rp_vltime;
1427                         prl->prefix[i].pltime = rpp->rp_pltime;
1428                         prl->prefix[i].if_index = rpp->rp_ifp->if_index;
1429                         prl->prefix[i].expire = rpp->rp_expire;
1430                         prl->prefix[i].advrtrs = 0;
1431                         prl->prefix[i].origin = rpp->rp_origin;
1432                         i++;
1433                 }
1434               }
1435                 splx(s);
1436
1437                 break;
1438         case OSIOCGIFINFO_IN6:
1439                 if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1440                         error = EINVAL;
1441                         break;
1442                 }
1443                 ndi->ndi.linkmtu = nd_ifinfo[ifp->if_index].linkmtu;
1444                 ndi->ndi.maxmtu = nd_ifinfo[ifp->if_index].maxmtu;
1445                 ndi->ndi.basereachable =
1446                     nd_ifinfo[ifp->if_index].basereachable;
1447                 ndi->ndi.reachable = nd_ifinfo[ifp->if_index].reachable;
1448                 ndi->ndi.retrans = nd_ifinfo[ifp->if_index].retrans;
1449                 ndi->ndi.flags = nd_ifinfo[ifp->if_index].flags;
1450                 ndi->ndi.recalctm = nd_ifinfo[ifp->if_index].recalctm;
1451                 ndi->ndi.chlim = nd_ifinfo[ifp->if_index].chlim;
1452                 ndi->ndi.receivedra = nd_ifinfo[ifp->if_index].receivedra;
1453                 break;
1454         case SIOCGIFINFO_IN6:
1455                 if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1456                         error = EINVAL;
1457                         break;
1458                 }
1459                 ndi->ndi = nd_ifinfo[ifp->if_index];
1460                 break;
1461         case SIOCSIFINFO_FLAGS:
1462                 /* XXX: almost all other fields of ndi->ndi is unused */
1463                 if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1464                         error = EINVAL;
1465                         break;
1466                 }
1467                 nd_ifinfo[ifp->if_index].flags = ndi->ndi.flags;
1468                 break;
1469         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1470                 /* flush default router list */
1471                 /*
1472                  * xxx sumikawa: should not delete route if default
1473                  * route equals to the top of default router list
1474                  */
1475                 bzero(&any, sizeof(any));
1476                 defrouter_delreq(&any, 0);
1477                 defrouter_select();
1478                 /* xxx sumikawa: flush prefix list */
1479                 break;
1480         case SIOCSPFXFLUSH_IN6:
1481             {
1482                 /* flush all the prefix advertised by routers */
1483                 struct nd_prefix *pr, *next;
1484
1485                 s = splnet();
1486                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1487                         struct in6_ifaddr *ia, *ia_next;
1488
1489                         next = pr->ndpr_next;
1490
1491                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1492                                 continue; /* XXX */
1493
1494                         /* do we really have to remove addresses as well? */
1495                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1496                                 /* ia might be removed.  keep the next ptr. */
1497                                 ia_next = ia->ia_next;
1498
1499                                 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1500                                         continue;
1501
1502                                 if (ia->ia6_ndpr == pr)
1503                                         in6_purgeaddr(&ia->ia_ifa);
1504                         }
1505                         prelist_remove(pr);
1506                 }
1507                 splx(s);
1508                 break;
1509             }
1510         case SIOCSRTRFLUSH_IN6:
1511             {
1512                 /* flush all the default routers */
1513                 struct nd_defrouter *dr, *next;
1514
1515                 s = splnet();
1516                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1517                         /*
1518                          * The first entry of the list may be stored in
1519                          * the routing table, so we'll delete it later.
1520                          */
1521                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1522                                 next = TAILQ_NEXT(dr, dr_entry);
1523                                 defrtrlist_del(dr);
1524                         }
1525                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1526                 }
1527                 splx(s);
1528                 break;
1529             }
1530         case SIOCGNBRINFO_IN6:
1531             {
1532                 struct llinfo_nd6 *ln;
1533                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1534
1535                 /*
1536                  * XXX: KAME specific hack for scoped addresses
1537                  *      XXXX: for other scopes than link-local?
1538                  */
1539                 if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1540                     IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1541                         u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1542
1543                         if (*idp == 0)
1544                                 *idp = htons(ifp->if_index);
1545                 }
1546
1547                 s = splnet();
1548                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1549                         error = EINVAL;
1550                         splx(s);
1551                         break;
1552                 }
1553                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1554                 nbi->state = ln->ln_state;
1555                 nbi->asked = ln->ln_asked;
1556                 nbi->isrouter = ln->ln_router;
1557                 nbi->expire = ln->ln_expire;
1558                 splx(s);
1559                 
1560                 break;
1561             }
1562         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1563                 ndif->ifindex = nd6_defifindex;
1564                 break;
1565         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1566                 return(nd6_setdefaultiface(ndif->ifindex));
1567                 break;
1568         }
1569         return(error);
1570 }
1571
1572 /*
1573  * Create neighbor cache entry and cache link-layer address,
1574  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1575  */
1576 struct rtentry *
1577 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1578                  int lladdrlen,
1579                  int type,      /* ICMP6 type */
1580                  int code       /* type dependent information */)
1581 {
1582         struct rtentry *rt = NULL;
1583         struct llinfo_nd6 *ln = NULL;
1584         int is_newentry;
1585         struct sockaddr_dl *sdl = NULL;
1586         int do_update;
1587         int olladdr;
1588         int llchange;
1589         int newstate = 0;
1590
1591         if (!ifp)
1592                 panic("ifp == NULL in nd6_cache_lladdr");
1593         if (!from)
1594                 panic("from == NULL in nd6_cache_lladdr");
1595
1596         /* nothing must be updated for unspecified address */
1597         if (IN6_IS_ADDR_UNSPECIFIED(from))
1598                 return NULL;
1599
1600         /*
1601          * Validation about ifp->if_addrlen and lladdrlen must be done in
1602          * the caller.
1603          *
1604          * XXX If the link does not have link-layer adderss, what should
1605          * we do? (ifp->if_addrlen == 0)
1606          * Spec says nothing in sections for RA, RS and NA.  There's small
1607          * description on it in NS section (RFC 2461 7.2.3).
1608          */
1609
1610         rt = nd6_lookup(from, 0, ifp);
1611         if (!rt) {
1612 #if 0
1613                 /* nothing must be done if there's no lladdr */
1614                 if (!lladdr || !lladdrlen)
1615                         return NULL;
1616 #endif
1617
1618                 rt = nd6_lookup(from, 1, ifp);
1619                 is_newentry = 1;
1620         } else {
1621                 /* do nothing if static ndp is set */
1622                 if (rt->rt_flags & RTF_STATIC)
1623                         return NULL;
1624                 is_newentry = 0;
1625         }
1626
1627         if (!rt)
1628                 return NULL;
1629         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1630 fail:
1631                 (void)nd6_free(rt);
1632                 return NULL;
1633         }
1634         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1635         if (!ln)
1636                 goto fail;
1637         if (!rt->rt_gateway)
1638                 goto fail;
1639         if (rt->rt_gateway->sa_family != AF_LINK)
1640                 goto fail;
1641         sdl = SDL(rt->rt_gateway);
1642
1643         olladdr = (sdl->sdl_alen) ? 1 : 0;
1644         if (olladdr && lladdr) {
1645                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1646                         llchange = 1;
1647                 else
1648                         llchange = 0;
1649         } else
1650                 llchange = 0;
1651
1652         /*
1653          * newentry olladdr  lladdr  llchange   (*=record)
1654          *      0       n       n       --      (1)
1655          *      0       y       n       --      (2)
1656          *      0       n       y       --      (3) * STALE
1657          *      0       y       y       n       (4) *
1658          *      0       y       y       y       (5) * STALE
1659          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1660          *      1       --      y       --      (7) * STALE
1661          */
1662
1663         if (lladdr) {           /* (3-5) and (7) */
1664                 /*
1665                  * Record source link-layer address
1666                  * XXX is it dependent to ifp->if_type?
1667                  */
1668                 sdl->sdl_alen = ifp->if_addrlen;
1669                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1670         }
1671
1672         if (!is_newentry) {
1673                 if ((!olladdr && lladdr)                /* (3) */
1674                  || (olladdr && lladdr && llchange)) {  /* (5) */
1675                         do_update = 1;
1676                         newstate = ND6_LLINFO_STALE;
1677                 } else                                  /* (1-2,4) */
1678                         do_update = 0;
1679         } else {
1680                 do_update = 1;
1681                 if (!lladdr)                            /* (6) */
1682                         newstate = ND6_LLINFO_NOSTATE;
1683                 else                                    /* (7) */
1684                         newstate = ND6_LLINFO_STALE;
1685         }
1686
1687         if (do_update) {
1688                 /*
1689                  * Update the state of the neighbor cache.
1690                  */
1691                 ln->ln_state = newstate;
1692
1693                 if (ln->ln_state == ND6_LLINFO_STALE) {
1694                         /*
1695                          * XXX: since nd6_output() below will cause
1696                          * state tansition to DELAY and reset the timer,
1697                          * we must set the timer now, although it is actually
1698                          * meaningless.
1699                          */
1700                         ln->ln_expire = time_second + nd6_gctimer;
1701
1702                         if (ln->ln_hold) {
1703                                 /*
1704                                  * we assume ifp is not a p2p here, so just
1705                                  * set the 2nd argument as the 1st one.
1706                                  */
1707                                 nd6_output(ifp, ifp, ln->ln_hold,
1708                                            (struct sockaddr_in6 *)rt_key(rt),
1709                                            rt);
1710                                 ln->ln_hold = NULL;
1711                         }
1712                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1713                         /* probe right away */
1714                         ln->ln_expire = time_second;
1715                 }
1716         }
1717
1718         /*
1719          * ICMP6 type dependent behavior.
1720          *
1721          * NS: clear IsRouter if new entry
1722          * RS: clear IsRouter
1723          * RA: set IsRouter if there's lladdr
1724          * redir: clear IsRouter if new entry
1725          *
1726          * RA case, (1):
1727          * The spec says that we must set IsRouter in the following cases:
1728          * - If lladdr exist, set IsRouter.  This means (1-5).
1729          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1730          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1731          * A quetion arises for (1) case.  (1) case has no lladdr in the
1732          * neighbor cache, this is similar to (6).
1733          * This case is rare but we figured that we MUST NOT set IsRouter.
1734          *
1735          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1736          *                                                      D R
1737          *      0       n       n       --      (1)     c   ?     s
1738          *      0       y       n       --      (2)     c   s     s
1739          *      0       n       y       --      (3)     c   s     s
1740          *      0       y       y       n       (4)     c   s     s
1741          *      0       y       y       y       (5)     c   s     s
1742          *      1       --      n       --      (6) c   c       c s
1743          *      1       --      y       --      (7) c   c   s   c s
1744          *
1745          *                                      (c=clear s=set)
1746          */
1747         switch (type & 0xff) {
1748         case ND_NEIGHBOR_SOLICIT:
1749                 /*
1750                  * New entry must have is_router flag cleared.
1751                  */
1752                 if (is_newentry)        /* (6-7) */
1753                         ln->ln_router = 0;
1754                 break;
1755         case ND_REDIRECT:
1756                 /*
1757                  * If the icmp is a redirect to a better router, always set the
1758                  * is_router flag. Otherwise, if the entry is newly created,
1759                  * clear the flag. [RFC 2461, sec 8.3]
1760                  */
1761                 if (code == ND_REDIRECT_ROUTER)
1762                         ln->ln_router = 1;
1763                 else if (is_newentry) /* (6-7) */
1764                         ln->ln_router = 0;
1765                 break;
1766         case ND_ROUTER_SOLICIT:
1767                 /*
1768                  * is_router flag must always be cleared.
1769                  */
1770                 ln->ln_router = 0;
1771                 break;
1772         case ND_ROUTER_ADVERT:
1773                 /*
1774                  * Mark an entry with lladdr as a router.
1775                  */
1776                 if ((!is_newentry && (olladdr || lladdr))       /* (2-5) */
1777                  || (is_newentry && lladdr)) {                  /* (7) */
1778                         ln->ln_router = 1;
1779                 }
1780                 break;
1781         }
1782
1783         /*
1784          * When the link-layer address of a router changes, select the
1785          * best router again.  In particular, when the neighbor entry is newly
1786          * created, it might affect the selection policy.
1787          * Question: can we restrict the first condition to the "is_newentry"
1788          * case?
1789          * XXX: when we hear an RA from a new router with the link-layer
1790          * address option, defrouter_select() is called twice, since
1791          * defrtrlist_update called the function as well.  However, I believe
1792          * we can compromise the overhead, since it only happens the first
1793          * time.
1794          * XXX: although defrouter_select() should not have a bad effect
1795          * for those are not autoconfigured hosts, we explicitly avoid such
1796          * cases for safety.
1797          */
1798         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1799                 defrouter_select();
1800
1801         return rt;
1802 }
1803
1804 static void
1805 nd6_slowtimo(void *ignored_arg)
1806 {
1807         int s = splnet();
1808         int i;
1809         struct nd_ifinfo *nd6if;
1810
1811         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1812             nd6_slowtimo, NULL);
1813         for (i = 1; i < if_index + 1; i++) {
1814                 if (!nd_ifinfo || i >= nd_ifinfo_indexlim)
1815                         continue;
1816                 nd6if = &nd_ifinfo[i];
1817                 if (nd6if->basereachable && /* already initialized */
1818                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1819                         /*
1820                          * Since reachable time rarely changes by router
1821                          * advertisements, we SHOULD insure that a new random
1822                          * value gets recomputed at least once every few hours.
1823                          * (RFC 2461, 6.3.4)
1824                          */
1825                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1826                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1827                 }
1828         }
1829         splx(s);
1830 }
1831
1832 #define senderr(e) { error = (e); goto bad;}
1833 int
1834 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1835            struct sockaddr_in6 *dst, struct rtentry *rt0)
1836 {
1837         struct mbuf *m = m0;
1838         struct rtentry *rt = rt0;
1839         struct sockaddr_in6 *gw6 = NULL;
1840         struct llinfo_nd6 *ln = NULL;
1841         int error = 0;
1842
1843         if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1844                 goto sendpkt;
1845
1846         if (nd6_need_cache(ifp) == 0)
1847                 goto sendpkt;
1848
1849         /*
1850          * next hop determination.  This routine is derived from ether_outpout.
1851          */
1852         if (rt) {
1853                 if (!(rt->rt_flags & RTF_UP)) {
1854                         if ((rt0 = rt =
1855                             rtlookup((struct sockaddr *)dst, 1, 0UL))) {
1856                                 rt->rt_refcnt--;
1857                                 if (rt->rt_ifp != ifp) {
1858                                         /* XXX: loop care? */
1859                                         return nd6_output(ifp, origifp, m0,
1860                                                           dst, rt);
1861                                 }
1862                         } else
1863                                 senderr(EHOSTUNREACH);
1864                 }
1865
1866                 if (rt->rt_flags & RTF_GATEWAY) {
1867                         gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1868
1869                         /*
1870                          * We skip link-layer address resolution and NUD
1871                          * if the gateway is not a neighbor from ND point
1872                          * of view, regardless of the value of nd_ifinfo.flags.
1873                          * The second condition is a bit tricky; we skip
1874                          * if the gateway is our own address, which is
1875                          * sometimes used to install a route to a p2p link.
1876                          */
1877                         if (!nd6_is_addr_neighbor(gw6, ifp) ||
1878                             in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1879                                 /*
1880                                  * We allow this kind of tricky route only
1881                                  * when the outgoing interface is p2p.
1882                                  * XXX: we may need a more generic rule here.
1883                                  */
1884                                 if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1885                                         senderr(EHOSTUNREACH);
1886
1887                                 goto sendpkt;
1888                         }
1889
1890                         if (rt->rt_gwroute == NULL)
1891                                 goto lookup;
1892                         if (!(rt->rt_gwroute->rt_flags & RTF_UP)) {
1893                                 rtfree(rt->rt_gwroute);
1894 lookup:                         rt->rt_gwroute = rtlookup(rt->rt_gateway, 1,
1895                                                           0UL);
1896                                 if (rt->rt_gwroute == NULL)
1897                                         senderr(EHOSTUNREACH);
1898                         }
1899                 }
1900         }
1901
1902         /*
1903          * Address resolution or Neighbor Unreachability Detection
1904          * for the next hop.
1905          * At this point, the destination of the packet must be a unicast
1906          * or an anycast address(i.e. not a multicast).
1907          */
1908
1909         /* Look up the neighbor cache for the nexthop */
1910         if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1911                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1912         else {
1913                 /*
1914                  * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1915                  * the condition below is not very efficient.  But we believe
1916                  * it is tolerable, because this should be a rare case.
1917                  */
1918                 if (nd6_is_addr_neighbor(dst, ifp) &&
1919                     (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1920                         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1921         }
1922         if (!ln || !rt) {
1923                 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1924                     !(nd_ifinfo[ifp->if_index].flags & ND6_IFF_PERFORMNUD)) {
1925                         log(LOG_DEBUG,
1926                             "nd6_output: can't allocate llinfo for %s "
1927                             "(ln=%p, rt=%p)\n",
1928                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1929                         senderr(EIO);   /* XXX: good error? */
1930                 }
1931
1932                 goto sendpkt;   /* send anyway */
1933         }
1934
1935         /* We don't have to do link-layer address resolution on a p2p link. */
1936         if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1937             ln->ln_state < ND6_LLINFO_REACHABLE) {
1938                 ln->ln_state = ND6_LLINFO_STALE;
1939                 ln->ln_expire = time_second + nd6_gctimer;
1940         }
1941
1942         /*
1943          * The first time we send a packet to a neighbor whose entry is
1944          * STALE, we have to change the state to DELAY and a sets a timer to
1945          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1946          * neighbor unreachability detection on expiration.
1947          * (RFC 2461 7.3.3)
1948          */
1949         if (ln->ln_state == ND6_LLINFO_STALE) {
1950                 ln->ln_asked = 0;
1951                 ln->ln_state = ND6_LLINFO_DELAY;
1952                 ln->ln_expire = time_second + nd6_delay;
1953         }
1954
1955         /*
1956          * If the neighbor cache entry has a state other than INCOMPLETE
1957          * (i.e. its link-layer address is already resolved), just
1958          * send the packet.
1959          */
1960         if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1961                 goto sendpkt;
1962
1963         /*
1964          * There is a neighbor cache entry, but no ethernet address
1965          * response yet.  Replace the held mbuf (if any) with this
1966          * latest one.
1967          *
1968          * This code conforms to the rate-limiting rule described in Section
1969          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1970          * an NS below.
1971          */
1972         if (ln->ln_state == ND6_LLINFO_NOSTATE)
1973                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1974         if (ln->ln_hold)
1975                 m_freem(ln->ln_hold);
1976         ln->ln_hold = m;
1977         if (ln->ln_expire) {
1978                 if (ln->ln_asked < nd6_mmaxtries &&
1979                     ln->ln_expire < time_second) {
1980                         ln->ln_asked++;
1981                         ln->ln_expire = time_second +
1982                                 nd_ifinfo[ifp->if_index].retrans / 1000;
1983                         nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1984                 }
1985         }
1986         return(0);
1987         
1988   sendpkt:
1989
1990         if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1991                 return((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1992                                          rt));
1993         }
1994         return((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
1995
1996   bad:
1997         if (m)
1998                 m_freem(m);
1999         return (error);
2000 }
2001 #undef senderr
2002
2003 int
2004 nd6_need_cache(struct ifnet *ifp)
2005 {
2006         /*
2007          * XXX: we currently do not make neighbor cache on any interface
2008          * other than ARCnet, Ethernet, FDDI and GIF.
2009          *
2010          * RFC2893 says:
2011          * - unidirectional tunnels needs no ND
2012          */
2013         switch (ifp->if_type) {
2014         case IFT_ARCNET:
2015         case IFT_ETHER:
2016         case IFT_FDDI:
2017         case IFT_IEEE1394:
2018 #ifdef IFT_L2VLAN
2019         case IFT_L2VLAN:
2020 #endif
2021 #ifdef IFT_IEEE80211
2022         case IFT_IEEE80211:
2023 #endif
2024         case IFT_GIF:           /* XXX need more cases? */
2025                 return(1);
2026         default:
2027                 return(0);
2028         }
2029 }
2030
2031 int
2032 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
2033                 struct sockaddr *dst, u_char *desten)
2034 {
2035         struct sockaddr_dl *sdl;
2036         struct rtentry *rt;
2037
2038
2039         if (m->m_flags & M_MCAST) {
2040                 switch (ifp->if_type) {
2041                 case IFT_ETHER:
2042                 case IFT_FDDI:
2043 #ifdef IFT_L2VLAN
2044         case IFT_L2VLAN:
2045 #endif
2046 #ifdef IFT_IEEE80211
2047                 case IFT_IEEE80211:
2048 #endif
2049                         ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2050                                                  desten);
2051                         return(1);
2052                 case IFT_IEEE1394:
2053                         bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen); 
2054                         return(1);
2055                 case IFT_ARCNET:
2056                         *desten = 0;
2057                         return(1);
2058                 default:
2059                         m_freem(m);
2060                         return(0);
2061                 }
2062         }
2063         if (rt0 == NULL) {
2064                 /* this could happen, if we could not allocate memory */
2065                 m_freem(m);
2066                 return(0);
2067         }
2068         if (rt_llroute(dst, rt0, &rt) != 0) {
2069                 m_freem(m);
2070                 return (0);
2071         }
2072         if (rt->rt_gateway->sa_family != AF_LINK) {
2073                 printf("nd6_storelladdr: something odd happens\n");
2074                 m_freem(m);
2075                 return(0);
2076         }
2077         sdl = SDL(rt->rt_gateway);
2078         if (sdl->sdl_alen == 0) {
2079                 /* this should be impossible, but we bark here for debugging */
2080                 printf("nd6_storelladdr: sdl_alen == 0\n");
2081                 m_freem(m);
2082                 return(0);
2083         }
2084
2085         bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2086         return(1);
2087 }
2088
2089 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2090 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2091 #ifdef SYSCTL_DECL
2092 SYSCTL_DECL(_net_inet6_icmp6);
2093 #endif
2094 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2095         CTLFLAG_RD, nd6_sysctl_drlist, "");
2096 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2097         CTLFLAG_RD, nd6_sysctl_prlist, "");
2098
2099 static int
2100 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2101 {
2102         int error;
2103         char buf[1024];
2104         struct in6_defrouter *d, *de;
2105         struct nd_defrouter *dr;
2106
2107         if (req->newptr)
2108                 return EPERM;
2109         error = 0;
2110
2111         for (dr = TAILQ_FIRST(&nd_defrouter);
2112              dr;
2113              dr = TAILQ_NEXT(dr, dr_entry)) {
2114                 d = (struct in6_defrouter *)buf;
2115                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2116
2117                 if (d + 1 <= de) {
2118                         bzero(d, sizeof(*d));
2119                         d->rtaddr.sin6_family = AF_INET6;
2120                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2121                         if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2122                             dr->ifp) != 0)
2123                                 log(LOG_ERR,
2124                                     "scope error in "
2125                                     "default router list (%s)\n",
2126                                     ip6_sprintf(&dr->rtaddr));
2127                         d->flags = dr->flags;
2128                         d->rtlifetime = dr->rtlifetime;
2129                         d->expire = dr->expire;
2130                         d->if_index = dr->ifp->if_index;
2131                 } else
2132                         panic("buffer too short");
2133
2134                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2135                 if (error)
2136                         break;
2137         }
2138         return error;
2139 }
2140
2141 static int
2142 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2143 {
2144         int error;
2145         char buf[1024];
2146         struct in6_prefix *p, *pe;
2147         struct nd_prefix *pr;
2148
2149         if (req->newptr)
2150                 return EPERM;
2151         error = 0;
2152
2153         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2154                 u_short advrtrs;
2155                 size_t advance;
2156                 struct sockaddr_in6 *sin6, *s6;
2157                 struct nd_pfxrouter *pfr;
2158
2159                 p = (struct in6_prefix *)buf;
2160                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2161
2162                 if (p + 1 <= pe) {
2163                         bzero(p, sizeof(*p));
2164                         sin6 = (struct sockaddr_in6 *)(p + 1);
2165
2166                         p->prefix = pr->ndpr_prefix;
2167                         if (in6_recoverscope(&p->prefix,
2168                             &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2169                                 log(LOG_ERR,
2170                                     "scope error in prefix list (%s)\n",
2171                                     ip6_sprintf(&p->prefix.sin6_addr));
2172                         p->raflags = pr->ndpr_raf;
2173                         p->prefixlen = pr->ndpr_plen;
2174                         p->vltime = pr->ndpr_vltime;
2175                         p->pltime = pr->ndpr_pltime;
2176                         p->if_index = pr->ndpr_ifp->if_index;
2177                         p->expire = pr->ndpr_expire;
2178                         p->refcnt = pr->ndpr_refcnt;
2179                         p->flags = pr->ndpr_stateflags;
2180                         p->origin = PR_ORIG_RA;
2181                         advrtrs = 0;
2182                         for (pfr = pr->ndpr_advrtrs.lh_first;
2183                              pfr;
2184                              pfr = pfr->pfr_next) {
2185                                 if ((void *)&sin6[advrtrs + 1] >
2186                                     (void *)pe) {
2187                                         advrtrs++;
2188                                         continue;
2189                                 }
2190                                 s6 = &sin6[advrtrs];
2191                                 bzero(s6, sizeof(*s6));
2192                                 s6->sin6_family = AF_INET6;
2193                                 s6->sin6_len = sizeof(*sin6);
2194                                 if (in6_recoverscope(s6,
2195                                     &pfr->router->rtaddr,
2196                                     pfr->router->ifp) != 0)
2197                                         log(LOG_ERR,
2198                                             "scope error in "
2199                                             "prefix list (%s)\n",
2200                                             ip6_sprintf(&pfr->router->rtaddr));
2201                                 advrtrs++;
2202                         }
2203                         p->advrtrs = advrtrs;
2204                 } else 
2205                         panic("buffer too short");
2206
2207                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2208                 error = SYSCTL_OUT(req, buf, advance);
2209                 if (error)
2210                         break;
2211         }
2212         return error;
2213 }